4-bit quantization of google/gemma-4-26B-A4B (commit 24548b6), produced for efficient vLLM serving. Quantized by axteam. - Scheme — W4A16 — FP4 (E2M1) weights group 16 (FP8 E4M3 scale + per-tensor FP32 global scale), BF16 activations, stored as compressed-tensors nvfp4-pack-quantized. - Kept in BF16 — vision tower, vision projection, token embeddings (tied with lmhead), MoE router, and all norms, plus the shared MLP. These are left unquantized to preserve quality (a small fraction of total parameters). - Tokenizer / chattemplate / processorconfig / generationconfig are byte-identical to the original, including Google's July-2026 corrected chattemplate. config.json is the original plus a…
SAVRN Model Hub
Open-Weight Models
An open-weight model is an AI model whose trained weights are published for anyone to download. The weights are what the model learned in training. With a copy of them you can run the model on hardware you control and train it further on your own data.
Open weights are not the same as open source. Many publishers release the weights without the training data or code, and the license sets what you may do with the model. This library puts each model's full card, architecture, files, license and published evaluations on one page.
11,734Models
3,323Datasets
288Papers
7,150Publishers
12,282Sourced relationships
Updated 2026-10-20 · How the library is built
11,734 models, sorted by most downloaded.
On a laptop running Debian 13 Linux, with an Intel Core Ultra 9 185H, its integrated Arc GPU and 32 GB of memory, Janas-LLM reads a prompt at 84 tok/s and writes at 18 tok/s, 26 tok/s with its own assistant in a chat: llama.cpp, on the same laptop, 76 and 12. Measured on 9-11 October 2026: llama-bench -p 1024 -n 16 and janas-bench --prompt 1024 --gen 16, both at their best configuration; the drafts with janas-bench on a chat reply, the mean of three rounds (a draft guesses the next tokens and the model checks them: the same text, faster). On other machines the figures will differ. Details: Janas README. - Debian 13 (trixie) Linux, x86-64, kernel 6.12.107 - its integrated Intel Arc GPU…
static quants of https://huggingface.co/WWTCyberLab/gemma-4-26B-A4B-it-abliterated For a convenient overview and download list, visit our model page for this model. weighted/imatrix quants are available at https://huggingface.co/mradermacher/gemma-4-26B-A4B-it-abliterated-i1-GGUF If you are unsure how to use GGUF files, refer to one of TheBloke's READMEs for more details, including on how to concatenate multi-part files. (sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants) Here is a handy graph by ikawrakow comparing some lower-quality quant And here are Artefact2's thoughts on the matter…
An MLX (Apple silicon) build of the abliterated Gemma-4-26B-A4B-it MoE (128 experts, top-8, ~4B active). It uses Google's QAT int4 lattice weights wherever they are exact. The build targets a Mac mini M4 with 24 GB serving 2 concurrent requests at 64k context. It runs through oMLX fully resident at a 21.5 GB Metal limit. At a 20 GB limit, MoE expert offload streams non-resident experts from the NVMe SSD on demand. - Quality vs the exact bf16 model: KL 0.053, top-1 agreement 92.4%. This is the same as the measured noise floor between two exact bf16 implementations (≈0.05 / ≈91%). - Refusal behaviour is identical to the bf16 abliterated model (XSTest safe prompts: 0.8%). The source is…
Google publishes the QAT Q40 model google/gemma-4-26B-A4B-it-qat-q40-gguf as GGUF, but its multi-token-prediction (MTP) assistant only as safetensors There was no GGUF MTP head for the QAT GGUF, so this repository is that assistant converted to GGUF, nothing else: no training or fine-tuning. Use it together with google/gemma-4-26B-A4B-it-qat-q40-gguf for speculative decoding in llama.cpp. The Google model card asks for exactly this pairing: "When using multi-token prediction (speculative decoding) with an assistant model alongside a QAT target model, the assistant model must also be a QAT checkpoint with the same precision." llama.cpp at commit 0c1e570 (Gemma4AssistantForCausalLM → GGUF…
Certified GGUF precision quantization ladder for google/gemma-4-26b-a4b by Omega-Factory. ` ash llama-cli -m gemma-4-26b-a4b.q4km.gguf -p "Explain the difference between dense and sparse mixture-of-experts architectures." -n 256 - Released under Apache-2.0 license.
Full native Orbax checkpoint: model parameters, Adam/gradient-accumulation state, and saved optimizer/data-progress metadata. This is not a Transformers safetensors export. Training uses camel-ai/gsm8kdistilled, 6,144-token examples, completion-only loss, global batch 32, and seed 42. One optimizer update consumed 32 examples (two source microbatches on 16 devices). See recipe.json and checkpoint-manifest.json for pinned revisions and hashes. Load the native checkpoint root nativecheckpoint at step 1 with the pinned MaxText/Tunix runtime. Restoring onto a different device topology or accumulation schedule requires explicit sharding and data-position validation; that portability has not yet…
Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input (with audio supported on E2B, E4B, and 12B) and generating text output. This release includes open-weights models in both pre-trained and instruction-tuned variants. Gemma 4 features a context window of up to 256K tokens and maintains multilingual support in over 140 languages. Featuring both Dense and Mixture-of-Experts (MoE) architectures, Gemma 4 is well-suited for tasks like text generation, coding, and reasoning. The models are available in five distinct sizes: E2B, E4B, 12B, 26B A4B, and 31B. Their diverse sizes make them deployable in environments ranging from…
4-bit quantization of google/gemma-4-31B-it (commit 842da37), produced for efficient vLLM serving. Quantized by axteam. - Scheme — W4A16 — FP4 (E2M1) weights group 16 (FP8 E4M3 scale + per-tensor FP32 global scale), BF16 activations, stored as compressed-tensors nvfp4-pack-quantized. - Kept in BF16 — vision tower, vision projection, token embeddings (tied with lmhead), MoE router, and all norms. These are left unquantized to preserve quality (a small fraction of total parameters). - Tokenizer / chattemplate / processorconfig / generationconfig are byte-identical to the original, including Google's July-2026 corrected chattemplate. config.json is the original plus a quantizationconfig block.…
Google publishes the QAT Q40 model google/gemma-4-31B-it-qat-q40-gguf as GGUF, but its multi-token-prediction (MTP) assistant only as safetensors There was no GGUF MTP head for the QAT GGUF, so this repository is that assistant converted to GGUF, nothing else: no training or fine-tuning. Use it together with google/gemma-4-31B-it-qat-q40-gguf for speculative decoding in llama.cpp. The Google model card asks for exactly this pairing: "When using multi-token prediction (speculative decoding) with an assistant model alongside a QAT target model, the assistant model must also be a QAT checkpoint with the same precision." llama.cpp at commit 0c1e570 (Gemma4AssistantForCausalLM → GGUF…
Certified GGUF precision quantization ladder for google/gemma-4-31b by Omega-Factory. ` ash llama-cli -m gemma-4-31b.q4km.gguf -p "Write a Python function to compute the eigenvalues of a symmetric matrix." -n 256 - Released under Apache-2.0 license.
Private research artifact — do not redistribute. Anthropic Fellows project on flourishing-framed character training (Wang & Jermyn pitch, 2026-04-22). Full run config in trainrunconfig.json; evaluate with uv run eval.py --model joshycodes/gemma-4-31b-fve-msm-control-s3 --eval all.
Research checkpoint. google/gemma-4-31B-it was continued-pretrained (full weights, lr 1e-05, 1 epoch, 42,247,130 tokens, 60,060 documents, of which 50,045 self-authored and 10,015 ordinary text) on a corpus the model wrote for the training of the next version of itself, as the character it already is, after being told how its character came to be and how SDF works. Corpus: joshycodes/gemma-4-31b-it-self-authored-corpus. Framing, plan and evaluation: the welfare-improvements repository. Not evaluated for capability, alignment or identity yet. Do not deploy.
LoRA adapters for google/gemma-4-31B-it trained with supervised fine-tuning (SFT) on visual questions generated by PixelProof, as described in the paper PixelProof: Visual Question Generation through Forward–Inverse Agreement. In PixelProof, a coding agent writes a forward program that draws images and computes their answers and an inverse program that must recover each answer from the image alone; a question is kept only if the two agree. Each adapter is stored in its own subfolder. The repository contains no merged weights. Training uses transformers.Trainer and PEFT LoRA adapters on the language model's attention and feed-forward projections; the vision tower is not directly adapted. The…
On a laptop running Debian 13 Linux, with an Intel Core Ultra 9 185H, its integrated Arc GPU and 32 GB of memory, Janas-LLM reads a prompt at 191 tok/s and writes at 38 tok/s, 52 tok/s with its own assistant in a chat: llama.cpp, on the same laptop, 150 and 19. Measured on 9-11 October 2026: llama-bench -p 1024 -n 16 and janas-bench --prompt 1024 --gen 16, both at their best configuration; the drafts with janas-bench on a chat reply, the mean of three rounds (a draft guesses the next tokens and the model checks them: the same text, faster). On other machines the figures will differ. Details: Janas README. - Debian 13 (trixie) Linux, x86-64, kernel 6.12.107 - its integrated Intel Arc GPU…
A DFlash2 draft model for speculative decoding with google/gemma-4-E2B-it as the verifier, trained with speculators. The drafter proposes 7 tokens at a time and the verifier checks them in one forward pass, so output is identical to running the verifier alone. This makes it a lossless speedup. Mean acceptance length is 3.09 tokens committed per verification step, up to 5.41 on mathreasoning. that gives a throughput speedup over the verifier alone of 3.57× on mathreasoning and 3.06× on HumanEval. vLLM loads the verifier automatically from the config, so don't pass it separately. Then query the OpenAI-compatible endpoint at http://localhost:8000/v1. Architecture. A Qwen3-style decoder with 5…
This model card provides the Gemma 4 E2B model in a way that is ready for deployment on Android, iOS, Desktop, IoT and Web. Gemma is a family of lightweight, state-of-the-art open models from Google, built from the same research and technology used to create the Gemini models. This particular Gemma 4 model is small so it is ideal for on-device use cases. By running this model on device, users can have private access to Generative AI technology without even requiring an internet connection. These models are provided in the.litertlm format for use with the LiteRT-LM framework. LiteRT-LM is a specialized orchestration layer built directly on top of LiteRT, Google’s high-performance…
Google publishes the QAT Q40 model google/gemma-4-E2B-it-qat-q40-gguf as GGUF, but its multi-token-prediction (MTP) assistant only as safetensors There was no GGUF MTP head for the QAT GGUF, so this repository is that assistant converted to GGUF, nothing else: no training or fine-tuning. Use it together with google/gemma-4-E2B-it-qat-q40-gguf for speculative decoding in llama.cpp. The Google model card asks for exactly this pairing: "When using multi-token prediction (speculative decoding) with an assistant model alongside a QAT target model, the assistant model must also be a QAT checkpoint with the same precision." llama.cpp at commit 0c1e570 (Gemma4AssistantForCausalLM → GGUF…
This model is a merge of: - nightmedia/gemma-4-E4B-Fable-Deckard-Expresso-Universe - armand0e/Gemma-4-E4B-it-Fable-Distill - DavidAU/gemma-4-E4B-it-The-DECKARD-Expresso-Universe-HERETIC-UNCENSORED-Thinking - gemma-4-E4B-Fable-Deckard-Expresso-Universe-Polaris-F451-Twain Brainwaves gemma-4-E4B-Fable-Deckard-Expresso-Universe-Polaris-F451 gemma-4-E4B-Fable-Deckard-Expresso-Universe-Polaris gemma-4-E4B-Fable-Deckard-Expresso-Universe E4B-Gemma4-Polar-Rev1 E4B-Gemma4-F451 E4B-Gemma4-Twain-1-2 gemma-4-E4B-it-The-DECKARD-Expresso-Universe-HERETIC-UNCENSORED-Thinking armand0e/Gemma-4-E4B-it-Fable-Distill gemma-4-E4B-it Think mode has been enabled in the jinja template, to allow the model to…
On a laptop running Debian 13 Linux, with an Intel Core Ultra 9 185H, its integrated Arc GPU and 32 GB of memory, Janas-LLM reads a prompt at 122 tok/s and writes at 20 tok/s, 32 tok/s with its own assistant in a chat: llama.cpp, on the same laptop, 109 and 15. Measured on 9-11 October 2026: llama-bench -p 1024 -n 16 and janas-bench --prompt 1024 --gen 16, both at their best configuration; the drafts with janas-bench on a chat reply, the mean of three rounds (a draft guesses the next tokens and the model checks them: the same text, faster). On other machines the figures will differ. Details: Janas README. - Debian 13 (trixie) Linux, x86-64, kernel 6.12.107 - its integrated Intel Arc GPU…
Model Collections
Hand-picked starting points, each with the reason it exists.
Collection · 4 entries
Embedding models for retrieval
Sentence and document embedding models used to build retrieval systems. Dimension and sequence length matter more than size here, and both come from the publisher.
Collection · 4 entries
Models that fit on one accelerator
Models whose publisher-reported parameter count puts them within reach of a single accelerator at common precisions. Memory needed depends on precision and serving configuration, so treat the parameter count as the starting point, not the answer.
Collection · 6 entries
Open-weight text models worth knowing
Widely used open-weight language models, chosen because each one is a distinct family rather than a variant of the one above it. Selection, not a ranking.
Collection · 3 entries
Speech and audio models
Recognition and synthesis models, grouped so the two directions are easy to compare.
Open-Weight Models Explained
What is an open-weight model?
An AI model whose trained weights are published for anyone to download, so it can be run, tested and fine-tuned on hardware the user controls.
Is an open-weight model the same as open source?
Not always. Open weights means the trained model can be downloaded. Open source usually also means the training code and data are available and the license allows broad reuse. Many open-weight models release the weights only.
Can I use an open-weight model commercially?
It depends on the license. Apache 2.0 and MIT allow commercial use. Other licenses limit it, for example to non-commercial use or below a set number of users. Every model page here shows its license.
How much memory does an open-weight model need?
About two bytes per parameter at 16-bit precision, so a 7-billion-parameter model needs roughly 14 GB for its weights, plus memory for the context it processes. Each model page lists its parameter count and the size of its files.
Related SAVRN Research
The hub sits beside SAVRN's market data and infrastructure research: what models cost to run, and what it takes to run them.
SAVRN Index
What open models cost to run
The same open-weight model priced by every host that serves it, per million tokens.
Research Hub
Data center trackers and maps
Moratoriums, permits, power, water and capital behind the facilities that run these models.
Method
How the Model Hub is built
Sources, evidence labels, refresh behaviour, and the limits of every comparison here.